diff options
Diffstat (limited to 'arch/ia64/sn/kernel')
| -rw-r--r-- | arch/ia64/sn/kernel/Makefile | 5 | ||||
| -rw-r--r-- | arch/ia64/sn/kernel/io_acpi_init.c | 234 | ||||
| -rw-r--r-- | arch/ia64/sn/kernel/io_common.c | 613 | ||||
| -rw-r--r-- | arch/ia64/sn/kernel/io_init.c | 633 | ||||
| -rw-r--r-- | arch/ia64/sn/kernel/iomv.c | 11 | ||||
| -rw-r--r-- | arch/ia64/sn/kernel/irq.c | 18 | ||||
| -rw-r--r-- | arch/ia64/sn/kernel/msi_sn.c | 4 | ||||
| -rw-r--r-- | arch/ia64/sn/kernel/setup.c | 38 | ||||
| -rw-r--r-- | arch/ia64/sn/kernel/sn2/sn_hwperf.c | 2 | ||||
| -rw-r--r-- | arch/ia64/sn/kernel/tiocx.c | 2 | ||||
| -rw-r--r-- | arch/ia64/sn/kernel/xpc_channel.c | 15 | ||||
| -rw-r--r-- | arch/ia64/sn/kernel/xpc_main.c | 64 |
12 files changed, 1068 insertions, 571 deletions
diff --git a/arch/ia64/sn/kernel/Makefile b/arch/ia64/sn/kernel/Makefile index 2d78f34dd763..0a59371d3475 100644 --- a/arch/ia64/sn/kernel/Makefile +++ b/arch/ia64/sn/kernel/Makefile @@ -4,13 +4,14 @@ # License. See the file "COPYING" in the main directory of this archive # for more details. # -# Copyright (C) 1999,2001-2005 Silicon Graphics, Inc. All Rights Reserved. +# Copyright (C) 1999,2001-2006 Silicon Graphics, Inc. All Rights Reserved. # CPPFLAGS += -I$(srctree)/arch/ia64/sn/include obj-y += setup.o bte.o bte_error.o irq.o mca.o idle.o \ - huberror.o io_init.o iomv.o klconflib.o pio_phys.o \ + huberror.o io_acpi_init.o io_common.o \ + io_init.o iomv.o klconflib.o pio_phys.o \ sn2/ obj-$(CONFIG_IA64_GENERIC) += machvec.o obj-$(CONFIG_SGI_TIOCX) += tiocx.o diff --git a/arch/ia64/sn/kernel/io_acpi_init.c b/arch/ia64/sn/kernel/io_acpi_init.c new file mode 100644 index 000000000000..cb96b4ea7df6 --- /dev/null +++ b/arch/ia64/sn/kernel/io_acpi_init.c @@ -0,0 +1,234 @@ +/* + * This file is subject to the terms and conditions of the GNU General Public + * License. See the file "COPYING" in the main directory of this archive + * for more details. + * + * Copyright (C) 2006 Silicon Graphics, Inc. All rights reserved. + */ + +#include <asm/sn/types.h> +#include <asm/sn/addrs.h> +#include <asm/sn/pcidev.h> +#include <asm/sn/pcibus_provider_defs.h> +#include <asm/sn/sn_sal.h> +#include "xtalk/hubdev.h" +#include <linux/acpi.h> + + +/* + * The code in this file will only be executed when running with + * a PROM that has ACPI IO support. (i.e., SN_ACPI_BASE_SUPPORT() == 1) + */ + + +/* + * This value must match the UUID the PROM uses + * (io/acpi/defblk.c) when building a vendor descriptor. + */ +struct acpi_vendor_uuid sn_uuid = { + .subtype = 0, + .data = { 0x2c, 0xc6, 0xa6, 0xfe, 0x9c, 0x44, 0xda, 0x11, + 0xa2, 0x7c, 0x08, 0x00, 0x69, 0x13, 0xea, 0x51 }, +}; + +/* + * Perform the early IO init in PROM. + */ +static s64 +sal_ioif_init(u64 *result) +{ + struct ia64_sal_retval isrv = {0,0,0,0}; + + SAL_CALL_NOLOCK(isrv, + SN_SAL_IOIF_INIT, 0, 0, 0, 0, 0, 0, 0); + *result = isrv.v0; + return isrv.status; +} + +/* + * sn_hubdev_add - The 'add' function of the acpi_sn_hubdev_driver. + * Called for every "SGIHUB" or "SGITIO" device defined + * in the ACPI namespace. + */ +static int __init +sn_hubdev_add(struct acpi_device *device) +{ + struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL }; + u64 addr; + struct hubdev_info *hubdev; + struct hubdev_info *hubdev_ptr; + int i; + u64 nasid; + struct acpi_resource *resource; + int ret = 0; + acpi_status status; + struct acpi_resource_vendor_typed *vendor; + extern void sn_common_hubdev_init(struct hubdev_info *); + + status = acpi_get_vendor_resource(device->handle, METHOD_NAME__CRS, + &sn_uuid, &buffer); + if (ACPI_FAILURE(status)) { + printk(KERN_ERR + "sn_hubdev_add: acpi_get_vendor_resource() failed: %d\n", + status); + return 1; + } + + resource = buffer.pointer; + vendor = &resource->data.vendor_typed; + if ((vendor->byte_length - sizeof(struct acpi_vendor_uuid)) != + sizeof(struct hubdev_info *)) { + printk(KERN_ERR + "sn_hubdev_add: Invalid vendor data length: %d\n", + vendor->byte_length); + ret = 1; + goto exit; + } + + memcpy(&addr, vendor->byte_data, sizeof(struct hubdev_info *)); + hubdev_ptr = __va((struct hubdev_info *) addr); + + nasid = hubdev_ptr->hdi_nasid; + i = nasid_to_cnodeid(nasid); + hubdev = (struct hubdev_info *)(NODEPDA(i)->pdinfo); + *hubdev = *hubdev_ptr; + sn_common_hubdev_init(hubdev); + +exit: + kfree(buffer.pointer); + return ret; +} + +/* + * sn_get_bussoft_ptr() - The pcibus_bussoft pointer is found in + * the ACPI Vendor resource for this bus. + */ +static struct pcibus_bussoft * +sn_get_bussoft_ptr(struct pci_bus *bus) +{ + u64 addr; + struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL }; + acpi_handle handle; + struct pcibus_bussoft *prom_bussoft_ptr; + struct acpi_resource *resource; + acpi_status status; + struct acpi_resource_vendor_typed *vendor; + + + handle = PCI_CONTROLLER(bus)->acpi_handle; + status = acpi_get_vendor_resource(handle, METHOD_NAME__CRS, + &sn_uuid, &buffer); + if (ACPI_FAILURE(status)) { + printk(KERN_ERR "get_acpi_pcibus_ptr: " + "get_acpi_bussoft_info() failed: %d\n", + status); + return NULL; + } + resource = buffer.pointer; + vendor = &resource->data.vendor_typed; + + if ((vendor->byte_length - sizeof(struct acpi_vendor_uuid)) != + sizeof(struct pcibus_bussoft *)) { + printk(KERN_ERR + "get_acpi_bussoft_ptr: Invalid vendor data " + "length %d\n", vendor->byte_length); + kfree(buffer.pointer); + return NULL; + } + memcpy(&addr, vendor->byte_data, sizeof(struct pcibus_bussoft *)); + prom_bussoft_ptr = __va((struct pcibus_bussoft *) addr); + kfree(buffer.pointer); + + return prom_bussoft_ptr; +} + +/* + * sn_acpi_bus_fixup + */ +void +sn_acpi_bus_fixup(struct pci_bus *bus) +{ + struct pci_dev *pci_dev = NULL; + struct pcibus_bussoft *prom_bussoft_ptr; + extern void sn_common_bus_fixup(struct pci_bus *, + struct pcibus_bussoft *); + + if (!bus->parent) { /* If root bus */ + prom_bussoft_ptr = sn_get_bussoft_ptr(bus); + if (prom_bussoft_ptr == NULL) { + printk(KERN_ERR + "sn_pci_fixup_bus: 0x%04x:0x%02x Unable to " + "obtain prom_bussoft_ptr\n", + pci_domain_nr(bus), bus->number); + return; + } + sn_common_bus_fixup(bus, prom_bussoft_ptr); + } + list_for_each_entry(pci_dev, &bus->devices, bus_list) { + sn_pci_fixup_slot(pci_dev); + } +} + +/* + * sn_acpi_slot_fixup - Perform any SN specific slot fixup. + * At present there does not appear to be + * any generic way to handle a ROM image + * that has been shadowed by the PROM, so + * we pass a pointer to it within the + * pcidev_info structure. + */ + +void +sn_acpi_slot_fixup(struct pci_dev *dev, struct pcidev_info *pcidev_info) +{ + void __iomem *addr; + size_t size; + + if (pcidev_info->pdi_pio_mapped_addr[PCI_ROM_RESOURCE]) { + /* + * A valid ROM image exists and has been shadowed by the + * PROM. Setup the pci_dev ROM resource to point to + * the shadowed copy. + */ + size = dev->resource[PCI_ROM_RESOURCE].end - + dev->resource[PCI_ROM_RESOURCE].start; + addr = + ioremap(pcidev_info->pdi_pio_mapped_addr[PCI_ROM_RESOURCE], + size); + dev->resource[PCI_ROM_RESOURCE].start = (unsigned long) addr; + dev->resource[PCI_ROM_RESOURCE].end = + (unsigned long) addr + size; + dev->resource[PCI_ROM_RESOURCE].flags |= IORESOURCE_ROM_BIOS_COPY; + } +} + +static struct acpi_driver acpi_sn_hubdev_driver = { + .name = "SGI HUBDEV Driver", + .ids = "SGIHUB,SGITIO", + .ops = { + .add = sn_hubdev_add, + }, +}; + + +/* + * sn_io_acpi_init - PROM has ACPI support for IO, defining at a minimum the + * nodes and root buses in the DSDT. As a result, bus scanning + * will be initiated by the Linux ACPI code. + */ + +void __init +sn_io_acpi_init(void) +{ + u64 result; + s64 status; + + /* SN Altix does not follow the IOSAPIC IRQ routing model */ + acpi_irq_model = ACPI_IRQ_MODEL_PLATFORM; + + acpi_bus_register_driver(&acpi_sn_hubdev_driver); + status = sal_ioif_init(&result); + if (status || result) + panic("sal_ioif_init failed: [%lx] %s\n", + status, ia64_sal_strerror(status)); +} diff --git a/arch/ia64/sn/kernel/io_common.c b/arch/ia64/sn/kernel/io_common.c new file mode 100644 index 000000000000..d4dd8f4b6b8d --- /dev/null +++ b/arch/ia64/sn/kernel/io_common.c @@ -0,0 +1,613 @@ +/* + * This file is subject to the terms and conditions of the GNU General Public + * License. See the file "COPYING" in the main directory of this archive + * for more details. + * + * Copyright (C) 2006 Silicon Graphics, Inc. All rights reserved. + */ + +#include <linux/bootmem.h> +#include <asm/sn/types.h> +#include <asm/sn/addrs.h> +#include <asm/sn/sn_feature_sets.h> +#include <asm/sn/geo.h> +#include <asm/sn/io.h> +#include <asm/sn/l1.h> +#include <asm/sn/module.h> +#include <asm/sn/pcibr_provider.h> +#include <asm/sn/pcibus_provider_defs.h> +#include <asm/sn/pcidev.h> +#include <asm/sn/simulator.h> +#include <asm/sn/sn_sal.h> +#include <asm/sn/tioca_provider.h> +#include <asm/sn/tioce_provider.h> +#include "xtalk/hubdev.h" +#include "xtalk/xwidgetdev.h" +#include <linux/acpi.h> +#include <asm/sn/sn2/sn_hwperf.h> +#include <asm/sn/acpi.h> + +extern void sn_init_cpei_timer(void); +extern void register_sn_procfs(void); +extern void sn_acpi_bus_fixup(struct pci_bus *); +extern void sn_bus_fixup(struct pci_bus *); +extern void sn_acpi_slot_fixup(struct pci_dev *, struct pcidev_info *); +extern void sn_more_slot_fixup(struct pci_dev *, struct pcidev_info *); +extern void sn_legacy_pci_window_fixup(struct pci_controller *, u64, u64); +extern void sn_io_acpi_init(void); +extern void sn_io_init(void); + + +static struct list_head sn_sysdata_list; + +/* sysdata list struct */ +struct sysdata_el { + struct list_head entry; + void *sysdata; +}; + +int sn_ioif_inited; /* SN I/O infrastructure initialized? */ + +struct sn_pcibus_provider *sn_pci_provider[PCIIO_ASIC_MAX_TYPES]; /* indexed by asic type */ + +/* + * Hooks and struct for unsupported pci providers + */ + +static dma_addr_t +sn_default_pci_map(struct pci_dev *pdev, unsigned long paddr, size_t size, int type) +{ + return 0; +} + +static void +sn_default_pci_unmap(struct pci_dev *pdev, dma_addr_t addr, int direction) +{ + return; +} + +static void * +sn_default_pci_bus_fixup(struct pcibus_bussoft *soft, struct pci_controller *controller) +{ + return NULL; +} + +static struct sn_pcibus_provider sn_pci_default_provider = { + .dma_map = sn_default_pci_map, + .dma_map_consistent = sn_default_pci_map, + .dma_unmap = sn_default_pci_unmap, + .bus_fixup = sn_default_pci_bus_fixup, +}; + +/* + * Retrieve the DMA Flush List given nasid, widget, and device. + * This list is needed to implement the WAR - Flush DMA data on PIO Reads. + */ +static inline u64 +sal_get_device_dmaflush_list(u64 nasid, u64 widget_num, u64 device_num, + u64 address) +{ + struct ia64_sal_retval ret_stuff; + ret_stuff.status = 0; + ret_stuff.v0 = 0; + + SAL_CALL_NOLOCK(ret_stuff, + (u64) SN_SAL_IOIF_GET_DEVICE_DMAFLUSH_LIST, + (u64) nasid, (u64) widget_num, + (u64) device_num, (u64) address, 0, 0, 0); + return ret_stuff.status; +} + +/* + * Retrieve the pci device information given the bus and device|function number. + */ +static inline u64 +sal_get_pcidev_info(u64 segment, u64 bus_number, u64 devfn, u64 pci_dev, + u64 sn_irq_info) +{ + struct ia64_sal_retval ret_stuff; + ret_stuff.status = 0; + ret_stuff.v0 = 0; + + SAL_CALL_NOLOCK(ret_stuff, + (u64) SN_SAL_IOIF_GET_PCIDEV_INFO, + (u64) segment, (u64) bus_number, (u64) devfn, + (u64) pci_dev, + sn_irq_info, 0, 0); + return ret_stuff.v0; +} + +/* + * sn_pcidev_info_get() - Retrieve the pcidev_info struct for the specified + * device. + */ +inline struct pcidev_info * +sn_pcidev_info_get(struct pci_dev *dev) +{ + struct pcidev_info *pcidev; + + list_for_each_entry(pcidev, + &(SN_PLATFORM_DATA(dev)->pcidev_info), pdi_list) { + if (pcidev->pdi_linux_pcidev == dev) + return pcidev; + } + return NULL; +} + +/* Older PROM flush WAR + * + * 01/16/06 -- This war will be in place until a new official PROM is released. + * Additionally note that the struct sn_flush_device_war also has to be + * removed from arch/ia64/sn/include/xtalk/hubdev.h + */ +static u8 war_implemented = 0; + +static s64 sn_device_fixup_war(u64 nasid, u64 widget, int device, + struct sn_flush_device_common *common) +{ + struct sn_flush_device_war *war_list; + struct sn_flush_device_war *dev_entry; + struct ia64_sal_retval isrv = {0,0,0,0}; + + if (!war_implemented) { + printk(KERN_WARNING "PROM version < 4.50 -- implementing old " + "PROM flush WAR\n"); + war_implemented = 1; + } + + war_list = kzalloc(DEV_PER_WIDGET * sizeof(*war_list), GFP_KERNEL); + if (!war_list) + BUG(); + + SAL_CALL_NOLOCK(isrv, SN_SAL_IOIF_GET_WIDGET_DMAFLUSH_LIST, + nasid, widget, __pa(war_list), 0, 0, 0 ,0); + if (isrv.status) + panic("sn_device_fixup_war failed: %s\n", + ia64_sal_strerror(isrv.status)); + + dev_entry = war_list + device; + memcpy(common,dev_entry, sizeof(*common)); + kfree(war_list); + + return isrv.status; +} + +/* + * sn_common_hubdev_init() - This routine is called to initialize the HUB data + * structure for each node in the system. + */ +void __init +sn_common_hubdev_init(struct hubdev_info *hubdev) +{ + + struct sn_flush_device_kernel *sn_flush_device_kernel; + struct sn_flush_device_kernel *dev_entry; + s64 status; + int widget, device, size; + + /* Attach the error interrupt handlers */ + if (hubdev->hdi_nasid & 1) /* If TIO */ + ice_error_init(hubdev); + else + hub_error_init(hubdev); + + for (widget = 0; widget <= HUB_WIDGET_ID_MAX; widget++) + hubdev->hdi_xwidget_info[widget].xwi_hubinfo = hubdev; + + if (!hubdev->hdi_flush_nasid_list.widget_p) + return; + + size = (HUB_WIDGET_ID_MAX + 1) * + sizeof(struct sn_flush_device_kernel *); + hubdev->hdi_flush_nasid_list.widget_p = + kzalloc(size, GFP_KERNEL); + if (!hubdev->hdi_flush_nasid_list.widget_p) + BUG(); + + for (widget = 0; widget <= HUB_WIDGET_ID_MAX; widget++) { + size = DEV_PER_WIDGET * + sizeof(struct sn_flush_device_kernel); + sn_flush_device_kernel = kzalloc(size, GFP_KERNEL); + if (!sn_flush_device_kernel) + BUG(); + + dev_entry = sn_flush_device_kernel; + for (device = 0; device < DEV_PER_WIDGET; + device++, dev_entry++) { + size = sizeof(struct sn_flush_device_common); + dev_entry->common = kzalloc(size, GFP_KERNEL); + if (!dev_entry->common) + BUG(); + if (sn_prom_feature_available(PRF_DEVICE_FLUSH_LIST)) + status = sal_get_device_dmaflush_list( + hubdev->hdi_nasid, widget, device, + (u64)(dev_entry->common)); + else + status = sn_device_fixup_war(hubdev->hdi_nasid, + widget, device, + dev_entry->common); + if (status != SALRET_OK) + panic("SAL call failed: %s\n", + ia64_sal_strerror(status)); + + spin_lock_init(&dev_entry->sfdl_flush_lock); + } + + if (sn_flush_device_kernel) + hubdev->hdi_flush_nasid_list.widget_p[widget] = + sn_flush_device_kernel; + } +} + +void sn_pci_unfixup_slot(struct pci_dev *dev) +{ + struct pci_dev *host_pci_dev = SN_PCIDEV_INFO(dev)->host_pci_dev; + + sn_irq_unfixup(dev); + pci_dev_put(host_pci_dev); + pci_dev_put(dev); +} + +/* + * sn_pci_fixup_slot() - This routine sets up a slot's resources consistent + * with the Linux PCI abstraction layer. Resources + * acquired from our PCI provider include PIO maps + * to BAR space and interrupt objects. + */ +void sn_pci_fixup_slot(struct pci_dev *dev) +{ + int segment = pci_domain_nr(dev->bus); + int status = 0; + struct pcibus_bussoft *bs; + struct pci_bus *host_pci_bus; + struct pci_dev *host_pci_dev; + struct pcidev_info *pcidev_info; + struct sn_irq_info *sn_irq_info; + unsigned int bus_no, devfn; + + pci_dev_get(dev); /* for the sysdata pointer */ + pcidev_info = kzalloc(sizeof(struct pcidev_info), GFP_KERNEL); + if (!pcidev_info) + BUG(); /* Cannot afford to run out of memory */ + + sn_irq_info = kzalloc(sizeof(struct sn_irq_info), GFP_KERNEL); + if (!sn_irq_info) + BUG(); /* Cannot afford to run out of memory */ + + /* Call to retrieve pci device information needed by kernel. */ + status = sal_get_pcidev_info((u64) segment, (u64) dev->bus->number, + dev->devfn, + (u64) __pa(pcidev_info), + (u64) __pa(sn_irq_info)); + if (status) + BUG(); /* Cannot get platform pci device information */ + + /* Add pcidev_info to list in pci_controller.platform_data */ + list_add_tail(&pcidev_info->pdi_list, + &(SN_PLATFORM_DATA(dev->bus)->pcidev_info)); + + if (SN_ACPI_BASE_SUPPORT()) + sn_acpi_slot_fixup(dev, pcidev_info); + else + sn_more_slot_fixup(dev, pcidev_info); + /* + * Using the PROMs values for the PCI host bus, get the Linux + * PCI host_pci_dev struct and set up host bus linkages + */ + + bus_no = (pcidev_info->pdi_slot_host_handle >> 32) & 0xff; + devfn = pcidev_info->pdi_slot_host_handle & 0xffffffff; + host_pci_bus = pci_find_bus(segment, bus_no); + host_pci_dev = pci_get_slot(host_pci_bus, devfn); + + pcidev_info->host_pci_dev = host_pci_dev; + pcidev_info->pdi_linux_pcidev = dev; + pcidev_info->pdi_host_pcidev_info = SN_PCIDEV_INFO(host_pci_dev); + bs = SN_PCIBUS_BUSSOFT(dev->bus); + pcidev_info->pdi_pcibus_info = bs; + + if (bs && bs->bs_asic_type < PCIIO_ASIC_MAX_TYPES) { + SN_PCIDEV_BUSPROVIDER(dev) = sn_pci_provider[bs->bs_asic_type]; + } else { + SN_PCIDEV_BUSPROVIDER(dev) = &sn_pci_default_provider; + } + + /* Only set up IRQ stuff if this device has a host bus context */ + if (bs && sn_irq_info->irq_irq) { + pcidev_info->pdi_sn_irq_info = sn_irq_info; + dev->irq = pcidev_info->pdi_sn_irq_info->irq_irq; + sn_irq_fixup(dev, sn_irq_info); + } else { + pcidev_info->pdi_sn_irq_info = NULL; + kfree(sn_irq_info); + } +} + +/* + * sn_common_bus_fixup - Perform platform specific bus fixup. + * Execute the ASIC specific fixup routine + * for this bus. + */ +void +sn_common_bus_fixup(struct pci_bus *bus, + struct pcibus_bussoft *prom_bussoft_ptr) +{ + int cnode; + struct pci_controller *controller; + struct hubdev_info *hubdev_info; + int nasid; + void *provider_soft; + struct sn_pcibus_provider *provider; + struct sn_platform_data *sn_platform_data; + + controller = PCI_CONTROLLER(bus); + /* + * Per-provider fixup. Copies the bus soft structure from prom + * to local area and links SN_PCIBUS_BUSSOFT(). + */ + + if (prom_bussoft_ptr->bs_asic_type >= PCIIO_ASIC_MAX_TYPES) { + printk(KERN_WARNING "sn_common_bus_fixup: Unsupported asic type, %d", + prom_bussoft_ptr->bs_asic_type); + return; + } + + if (prom_bussoft_ptr->bs_asic_type == PCIIO_ASIC_TYPE_PPB) + return; /* no further fixup necessary */ + + provider = sn_pci_provider[prom_bussoft_ptr->bs_asic_type]; + if (provider == NULL) + panic("sn_common_bus_fixup: No provider registered for this asic type, %d", + prom_bussoft_ptr->bs_asic_type); + + if (provider->bus_fixup) + provider_soft = (*provider->bus_fixup) (prom_bussoft_ptr, + controller); + else + provider_soft = NULL; + + /* + * Generic bus fixup goes here. Don't reference prom_bussoft_ptr + * after this point. + */ + controller->platform_data = kzalloc(sizeof(struct sn_platform_data), + GFP_KERNEL); + if (controller->platform_data == NULL) + BUG(); + sn_platform_data = + (struct sn_platform_data *) controller->platform_data; + sn_platform_data->provider_soft = provider_soft; + INIT_LIST_HEAD(&((struct sn_platform_data *) + controller->platform_data)->pcidev_info); + nasid = NASID_GET(SN_PCIBUS_BUSSOFT(bus)->bs_base); + cnode = nasid_to_cnodeid(nasid); + hubdev_info = (struct hubdev_info *)(NODEPDA(cnode)->pdinfo); + SN_PCIBUS_BUSSOFT(bus)->bs_xwidget_info = + &(hubdev_info->hdi_xwidget_info[SN_PCIBUS_BUSSOFT(bus)->bs_xid]); + + /* + * If the node information we obtained during the fixup phase is + * invalid then set controller->node to -1 (undetermined) + */ + if (controller->node >= num_online_nodes()) { + struct pcibus_bussoft *b = SN_PCIBUS_BUSSOFT(bus); + + printk(KERN_WARNING "Device ASIC=%u XID=%u PBUSNUM=%u" + "L_IO=%lx L_MEM=%lx BASE=%lx\n", + b->bs_asic_type, b->bs_xid, b->bs_persist_busnum, + b->bs_legacy_io, b->bs_legacy_mem, b->bs_base); + printk(KERN_WARNING "on node %d but only %d nodes online." + "Association set to undetermined.\n", + controller->node, num_online_nodes()); + controller->node = -1; + } +} + +void sn_bus_store_sysdata(struct pci_dev *dev) +{ + struct sysdata_el *element; + + element = kzalloc(sizeof(struct sysdata_el), GFP_KERNEL); + if (!element) { + dev_dbg(dev, "%s: out of memory!\n", __FUNCTION__); + return; + } + element->sysdata = SN_PCIDEV_INFO(dev); + list_add(&element->entry, &sn_sysdata_list); +} + +void sn_bus_free_sysdata(void) +{ + struct sysdata_el *element; + struct list_head *list, *safe; + + list_for_each_safe(list, safe, &sn_sysdata_list) { + element = list_entry(list, struct sysdata_el, entry); + list_del(&element->entry); + list_del(&(((struct pcidev_info *) + (element->sysdata))->pdi_list)); + kfree(element->sysdata); + kfree(element); + } + return; +} + +/* + * hubdev_init_node() - Creates the HUB data structure and link them to it's + * own NODE specific data area. + */ +void hubdev_init_node(nodepda_t * npda, cnodeid_t node) +{ + struct hubdev_info *hubdev_info; + int size; + pg_data_t *pg; + + size = sizeof(struct hubdev_info); + + if (node >= num_online_nodes()) /* Headless/memless IO nodes */ + pg = NODE_DATA(0); + else + pg = NODE_DATA(node); + + hubdev_info = (struct hubdev_info *)alloc_bootmem_node(pg, size); + + npda->pdinfo = (void *)hubdev_info; +} + +geoid_t +cnodeid_get_geoid(cnodeid_t cnode) +{ + struct hubdev_info *hubdev; + + hubdev = (struct hubdev_info *)(NODEPDA(cnode)->pdinfo); + return hubdev->hdi_geoid; +} + +void sn_generate_path(struct pci_bus *pci_bus, char *address) +{ + nasid_t nasid; + cnodeid_t cnode; + geoid_t geoid; + moduleid_t moduleid; + u16 bricktype; + + nasid = NASID_GET(SN_PCIBUS_BUSSOFT(pci_bus)->bs_base); + cnode = nasid_to_cnodeid(nasid); + geoid = cnodeid_get_geoid(cnode); + moduleid = geo_module(geoid); + + sprintf(address, "module_%c%c%c%c%.2d", + '0'+RACK_GET_CLASS(MODULE_GET_RACK(moduleid)), + '0'+RACK_GET_GROUP(MODULE_GET_RACK(moduleid)), + '0'+RACK_GET_NUM(MODULE_GET_RACK(moduleid)), + MODULE_GET_BTCHAR(moduleid), MODULE_GET_BPOS(moduleid)); + + /* Tollhouse requires slot id to be displayed */ + bricktype = MODULE_GET_BTYPE(moduleid); + if ((bricktype == L1_BRICKTYPE_191010) || + (bricktype == L1_BRICKTYPE_1932)) + sprintf(address, "%s^%d", address, geo_slot(geoid)); +} + +/* + * sn_pci_fixup_bus() - Perform SN specific setup of software structs + * (pcibus_bussoft, pcidev_info) and hardware + * registers, for the specified bus and devices under it. + */ +void __devinit +sn_pci_fixup_bus(struct pci_bus *bus) +{ + + if (SN_ACPI_BASE_SUPPORT()) + sn_acpi_bus_fixup(bus); + else + sn_bus_fixup(bus); +} + +/* + * sn_io_early_init - Perform early IO (and some non-IO) initialization. + * In particular, setup the sn_pci_provider[] array. + * This needs to be done prior to any bus scanning + * (acpi_scan_init()) in the ACPI case, as the SN + * bus fixup code will reference the array. + */ +static int __init +sn_io_early_init(void) +{ + int i; + + if (!ia64_platform_is("sn2") || IS_RUNNING_ON_FAKE_PROM()) + return 0; + + /* + * prime sn_pci_provider[]. Individial provider init routines will + * override their respective default entries. + */ + + for (i = 0; i < PCIIO_ASIC_MAX_TYPES; i++) + sn_pci_provider[i] = &sn_pci_default_provider; + + pcibr_init_provider(); + tioca_init_provider(); + tioce_init_provider(); + + /* + * This is needed to avoid bounce limit checks in the blk layer + */ + ia64_max_iommu_merge_mask = ~PAGE_MASK; + + sn_irq_lh_init(); + INIT_LIST_HEAD(&sn_sysdata_list); + sn_init_cpei_timer(); + +#ifdef CONFIG_PROC_FS + register_sn_procfs(); +#endif + + printk(KERN_INFO "ACPI DSDT OEM Rev 0x%x\n", + acpi_gbl_DSDT->oem_revision); + if (SN_ACPI_BASE_SUPPORT()) + sn_io_acpi_init(); + else + sn_io_init(); + return 0; +} + +arch_initcall(sn_io_early_init); + +/* + * sn_io_late_init() - Perform any final platform specific IO initialization. + */ + +int __init +sn_io_late_init(void) +{ + struct pci_bus *bus; + struct pcibus_bussoft *bussoft; + cnodeid_t cnode; + nasid_t nasid; + cnodeid_t near_cnode; + + if (!ia64_platform_is("sn2") || IS_RUNNING_ON_FAKE_PROM()) + return 0; + + /* + * Setup closest node in pci_controller->node for + * PIC, TIOCP, TIOCE (TIOCA does it during bus fixup using + * info from the PROM). + */ + bus = NULL; + while ((bus = pci_find_next_bus(bus)) != NULL) { + bussoft = SN_PCIBUS_BUSSOFT(bus); + nasid = NASID_GET(bussoft->bs_base); + cnode = nasid_to_cnodeid(nasid); + if ((bussoft->bs_asic_type == PCIIO_ASIC_TYPE_TIOCP) || + (bussoft->bs_asic_type == PCIIO_ASIC_TYPE_TIOCE)) { + /* TIO PCI Bridge: find nearest node with CPUs */ + int e = sn_hwperf_get_nearest_node(cnode, NULL, + &near_cnode); + if (e < 0) { + near_cnode = (cnodeid_t)-1; /* use any node */ + printk(KERN_WARNING "pcibr_bus_fixup: failed " + "to find near node with CPUs to TIO " + "node %d, err=%d\n", cnode, e); + } + PCI_CONTROLLER(bus)->node = near_cnode; + } else if (bussoft->bs_asic_type == PCIIO_ASIC_TYPE_PIC) { + PCI_CONTROLLER(bus)->node = cnode; + } + } + + sn_ioif_inited = 1; /* SN I/O infrastructure now initialized */ + + return 0; +} + +fs_initcall(sn_io_late_init); + +EXPORT_SYMBOL(sn_pci_fixup_slot); +EXPORT_SYMBOL(sn_pci_unfixup_slot); +EXPORT_SYMBOL(sn_bus_store_sysdata); +EXPORT_SYMBOL(sn_bus_free_sysdata); +EXPORT_SYMBOL(sn_generate_path); + diff --git a/arch/ia64/sn/kernel/io_init.c b/arch/ia64/sn/kernel/io_init.c index dc09a6a28a37..9ad843e0383b 100644 --- a/arch/ia64/sn/kernel/io_init.c +++ b/arch/ia64/sn/kernel/io_init.c @@ -3,103 +3,28 @@ * License. See the file "COPYING" in the main directory of this archive * for more details. * - * Copyright (C) 1992 - 1997, 2000-2005 Silicon Graphics, Inc. All rights reserved. + * Copyright (C) 1992 - 1997, 2000-2006 Silicon Graphics, Inc. All rights reserved. */ -#include <linux/bootmem.h> -#include <linux/nodemask.h> #include <asm/sn/types.h> #include <asm/sn/addrs.h> -#include <asm/sn/sn_feature_sets.h> -#include <asm/sn/geo.h> #include <asm/sn/io.h> -#include <asm/sn/l1.h> #include <asm/sn/module.h> -#include <asm/sn/pcibr_provider.h> +#include <asm/sn/intr.h> #include <asm/sn/pcibus_provider_defs.h> #include <asm/sn/pcidev.h> -#include <asm/sn/simulator.h> #include <asm/sn/sn_sal.h> -#include <asm/sn/tioca_provider.h> -#include <asm/sn/tioce_provider.h> #include "xtalk/hubdev.h" -#include "xtalk/xwidgetdev.h" - - -extern void sn_init_cpei_timer(void); -extern void register_sn_procfs(void); - -static struct list_head sn_sysdata_list; - -/* sysdata list struct */ -struct sysdata_el { - struct list_head entry; - void *sysdata; -}; - -struct slab_info { - struct hubdev_info hubdev; -}; - -struct brick { - moduleid_t id; /* Module ID of this module */ - struct slab_info slab_info[MAX_SLABS + 1]; -}; - -int sn_ioif_inited; /* SN I/O infrastructure initialized? */ - -struct sn_pcibus_provider *sn_pci_provider[PCIIO_ASIC_MAX_TYPES]; /* indexed by asic type */ - -static int max_segment_number; /* Default highest segment number */ -static int max_pcibus_number = 255; /* Default highest pci bus number */ /* - * Hooks and struct for unsupported pci providers + * The code in this file will only be executed when running with + * a PROM that does _not_ have base ACPI IO support. + * (i.e., SN_ACPI_BASE_SUPPORT() == 0) */ -static dma_addr_t -sn_default_pci_map(struct pci_dev *pdev, unsigned long paddr, size_t size, int type) -{ - return 0; -} - -static void -sn_default_pci_unmap(struct pci_dev *pdev, dma_addr_t addr, int direction) -{ - return; -} - -static void * -sn_default_pci_bus_fixup(struct pcibus_bussoft *soft, struct pci_controller *controller) -{ - return NULL; -} - -static struct sn_pcibus_provider sn_pci_default_provider = { - .dma_map = sn_default_pci_map, - .dma_map_consistent = sn_default_pci_map, - .dma_unmap = sn_default_pci_unmap, - .bus_fixup = sn_default_pci_bus_fixup, -}; - -/* - * Retrieve the DMA Flush List given nasid, widget, and device. - * This list is needed to implement the WAR - Flush DMA data on PIO Reads. - */ -static inline u64 -sal_get_device_dmaflush_list(u64 nasid, u64 widget_num, u64 device_num, - u64 address) -{ - struct ia64_sal_retval ret_stuff; - ret_stuff.status = 0; - ret_stuff.v0 = 0; +static int max_segment_number; /* Default highest segment number */ +static int max_pcibus_number = 255; /* Default highest pci bus number */ - SAL_CALL_NOLOCK(ret_stuff, - (u64) SN_SAL_IOIF_GET_DEVICE_DMAFLUSH_LIST, - (u64) nasid, (u64) widget_num, - (u64) device_num, (u64) address, 0, 0, 0); - return ret_stuff.status; -} /* * Retrieve the hub device info structure for the given nasid. @@ -131,93 +56,20 @@ static inline u64 sal_get_pcibus_info(u64 segment, u64 busnum, u64 address) return ret_stuff.v0; } -/* - * Retrieve the pci device information given the bus and device|function number. - */ -static inline u64 -sal_get_pcidev_info(u64 segment, u64 bus_number, u64 devfn, u64 pci_dev, - u64 sn_irq_info) -{ - struct ia64_sal_retval ret_stuff; - ret_stuff.status = 0; - ret_stuff.v0 = 0; - - SAL_CALL_NOLOCK(ret_stuff, - (u64) SN_SAL_IOIF_GET_PCIDEV_INFO, - (u64) segment, (u64) bus_number, (u64) devfn, - (u64) pci_dev, - sn_irq_info, 0, 0); - return ret_stuff.v0; -} - -/* - * sn_pcidev_info_get() - Retrieve the pcidev_info struct for the specified - * device. - */ -inline struct pcidev_info * -sn_pcidev_info_get(struct pci_dev *dev) -{ - struct pcidev_info *pcidev; - - list_for_each_entry(pcidev, - &(SN_PCI_CONTROLLER(dev)->pcidev_info), pdi_list) { - if (pcidev->pdi_linux_pcidev == dev) { - return pcidev; - } - } - return NULL; -} - -/* Older PROM flush WAR - * - * 01/16/06 -- This war will be in place until a new official PROM is released. - * Additionally note that the struct sn_flush_device_war also has to be - * removed from arch/ia64/sn/include/xtalk/hubdev.h - */ -static u8 war_implemented = 0; - -static s64 sn_device_fixup_war(u64 nasid, u64 widget, int device, - struct sn_flush_device_common *common) -{ - struct sn_flush_device_war *war_list; - struct sn_flush_device_war *dev_entry; - struct ia64_sal_retval isrv = {0,0,0,0}; - - if (!war_implemented) { - printk(KERN_WARNING "PROM version < 4.50 -- implementing old " - "PROM flush WAR\n"); - war_implemented = 1; - } - - war_list = kzalloc(DEV_PER_WIDGET * sizeof(*war_list), GFP_KERNEL); - if (!war_list) - BUG(); - - SAL_CALL_NOLOCK(isrv, SN_SAL_IOIF_GET_WIDGET_DMAFLUSH_LIST, - nasid, widget, __pa(war_list), 0, 0, 0 ,0); - if (isrv.status) - panic("sn_device_fixup_war failed: %s\n", - ia64_sal_strerror(isrv.status)); - - dev_entry = war_list + device; - memcpy(common,dev_entry, sizeof(*common)); - kfree(war_list); - - return isrv.status; -} /* - * sn_fixup_ionodes() - This routine initializes the HUB data strcuture for - * each node in the system. + * sn_fixup_ionodes() - This routine initializes the HUB data structure for + * each node in the system. This function is only + * executed when running with a non-ACPI capable PROM. */ static void __init sn_fixup_ionodes(void) { - struct sn_flush_device_kernel *sn_flush_device_kernel; - struct sn_flush_device_kernel *dev_entry; + struct hubdev_info *hubdev; u64 status; u64 nasid; - int i, widget, device, size; + int i; + extern void sn_common_hubdev_init(struct hubdev_info *); /* * Get SGI Specific HUB chipset information. @@ -240,70 +92,47 @@ static void __init sn_fixup_ionodes(void) max_segment_number = hubdev->max_segment_number; max_pcibus_number = hubdev->max_pcibus_number; } + sn_common_hubdev_init(hubdev); + } +} - /* Attach the error interrupt handlers */ - if (nasid & 1) - ice_error_init(hubdev); - else - hub_error_init(hubdev); - - for (widget = 0; widget <= HUB_WIDGET_ID_MAX; widget++) - hubdev->hdi_xwidget_info[widget].xwi_hubinfo = hubdev; - - if (!hubdev->hdi_flush_nasid_list.widget_p) - continue; - - size = (HUB_WIDGET_ID_MAX + 1) * - sizeof(struct sn_flush_device_kernel *); - hubdev->hdi_flush_nasid_list.widget_p = - kzalloc(size, GFP_KERNEL); - if (!hubdev->hdi_flush_nasid_list.widget_p) +/* + * sn_pci_legacy_window_fixup - Create PCI controller windows for + * legacy IO and MEM space. This needs to + * be done here, as the PROM does not have + * ACPI support defining the root buses + * and their resources (_CRS), + */ +static void +sn_legacy_pci_window_fixup(struct pci_controller *controller, + u64 legacy_io, u64 legacy_mem) +{ + controller->window = kcalloc(2, sizeof(struct pci_window), + GFP_KERNEL); + if (controller->window == NULL) BUG(); - - for (widget = 0; widget <= HUB_WIDGET_ID_MAX; widget++) { - size = DEV_PER_WIDGET * - sizeof(struct sn_flush_device_kernel); - sn_flush_device_kernel = kzalloc(size, GFP_KERNEL); - if (!sn_flush_device_kernel) - BUG(); - - dev_entry = sn_flush_device_kernel; - for (device = 0; device < DEV_PER_WIDGET; - device++,dev_entry++) { - size = sizeof(struct sn_flush_device_common); - dev_entry->common = kzalloc(size, GFP_KERNEL); - if (!dev_entry->common) - BUG(); - - if (sn_prom_feature_available( - PRF_DEVICE_FLUSH_LIST)) - status = sal_get_device_dmaflush_list( - nasid, widget, device, - (u64)(dev_entry->common)); - else - status = sn_device_fixup_war(nasid, - widget, device, - dev_entry->common); - if (status != SALRET_OK) - panic("SAL call failed: %s\n", - ia64_sal_strerror(status)); - - spin_lock_init(&dev_entry->sfdl_flush_lock); - } - - if (sn_flush_device_kernel) - hubdev->hdi_flush_nasid_list.widget_p[widget] = - sn_flush_device_kernel; - } - } + controller->window[0].offset = legacy_io; + controller->window[0].resource.name = "legacy_io"; + controller->window[0].resource.flags = IORESOURCE_IO; + controller->window[0].resource.start = legacy_io; + controller->window[0].resource.end = + controller->window[0].resource.start + 0xffff; + controller->window[0].resource.parent = &ioport_resource; + controller->window[1].offset = legacy_mem; + controller->window[1].resource.name = "legacy_mem"; + controller->window[1].resource.flags = IORESOURCE_MEM; + controller->window[1].resource.start = legacy_mem; + controller->window[1].resource.end = + controller->window[1].resource.start + (1024 * 1024) - 1; + controller->window[1].resource.parent = &iomem_resource; + controller->windows = 2; } /* * sn_pci_window_fixup() - Create a pci_window for each device resource. - * Until ACPI support is added, we need this code - * to setup pci_windows for use by - * pcibios_bus_to_resource(), - * pcibios_resource_to_bus(), etc. + * It will setup pci_windows for use by + * pcibios_bus_to_resource(), pcibios_resource_to_bus(), + * etc. */ static void sn_pci_window_fixup(struct pci_dev *dev, unsigned int count, @@ -342,60 +171,22 @@ sn_pci_window_fixup(struct pci_dev *dev, unsigned int count, controller->window = new_window; } -void sn_pci_unfixup_slot(struct pci_dev *dev) -{ - struct pci_dev *host_pci_dev = SN_PCIDEV_INFO(dev)->host_pci_dev; - - sn_irq_unfixup(dev); - pci_dev_put(host_pci_dev); - pci_dev_put(dev); -} - /* - * sn_pci_fixup_slot() - This routine sets up a slot's resources - * consistent with the Linux PCI abstraction layer. Resources acquired - * from our PCI provider include PIO maps to BAR space and interrupt - * objects. + * sn_more_slot_fixup() - We are not running with an ACPI capable PROM, + * and need to convert the pci_dev->resource + * 'start' and 'end' addresses to mapped addresses, + * and setup the pci_controller->window array entries. */ -void sn_pci_fixup_slot(struct pci_dev *dev) +void +sn_more_slot_fixup(struct pci_dev *dev, struct pcidev_info *pcidev_info) { unsigned int count = 0; int idx; - int segment = pci_domain_nr(dev->bus); - int status = 0; - struct pcibus_bussoft *bs; - struct pci_bus *host_pci_bus; - struct pci_dev *host_pci_dev; - struct pcidev_info *pcidev_info; s64 pci_addrs[PCI_ROM_RESOURCE + 1]; - struct sn_irq_info *sn_irq_info; - unsigned long size; - unsigned int bus_no, devfn; - - pci_dev_get(dev); /* for the sysdata pointer */ - pcidev_info = kzalloc(sizeof(struct pcidev_info), GFP_KERNEL); - if (!pcidev_info) - BUG(); /* Cannot afford to run out of memory */ - - sn_irq_info = kzalloc(sizeof(struct sn_irq_info), GFP_KERNEL); - if (!sn_irq_info) - BUG(); /* Cannot afford to run out of memory */ - - /* Call to retrieve pci device information needed by kernel. */ - status = sal_get_pcidev_info((u64) segment, (u64) dev->bus->number, - dev->devfn, - (u64) __pa(pcidev_info), - (u64) __pa(sn_irq_info)); - if (status) - BUG(); /* Cannot get platform pci device information */ - - /* Add pcidev_info to list in sn_pci_controller struct */ - list_add_tail(&pcidev_info->pdi_list, - &(SN_PCI_CONTROLLER(dev->bus)->pcidev_info)); + unsigned long addr, end, size, start; /* Copy over PIO Mapped Addresses */ for (idx = 0; idx <= PCI_ROM_RESOURCE; idx++) { - unsigned long start, end, addr; if (!pcidev_info->pdi_pio_mapped_addr[idx]) { pci_addrs[idx] = -1; @@ -419,60 +210,28 @@ void sn_pci_fixup_slot(struct pci_dev *dev) dev->resource[idx].parent = &ioport_resource; else dev->resource[idx].parent = &iomem_resource; + /* If ROM, mark as shadowed in PROM */ + if (idx == PCI_ROM_RESOURCE) + dev->resource[idx].flags |= IORESOURCE_ROM_BIOS_COPY; } /* Create a pci_window in the pci_controller struct for * each device resource. */ if (count > 0) sn_pci_window_fixup(dev, count, pci_addrs); - - /* - * Using the PROMs values for the PCI host bus, get the Linux - * PCI host_pci_dev struct and set up host bus linkages - */ - - bus_no = (pcidev_info->pdi_slot_host_handle >> 32) & 0xff; - devfn = pcidev_info->pdi_slot_host_handle & 0xffffffff; - host_pci_bus = pci_find_bus(segment, bus_no); - host_pci_dev = pci_get_slot(host_pci_bus, devfn); - - pcidev_info->host_pci_dev = host_pci_dev; - pcidev_info->pdi_linux_pcidev = dev; - pcidev_info->pdi_host_pcidev_info = SN_PCIDEV_INFO(host_pci_dev); - bs = SN_PCIBUS_BUSSOFT(dev->bus); - pcidev_info->pdi_pcibus_info = bs; - - if (bs && bs->bs_asic_type < PCIIO_ASIC_MAX_TYPES) { - SN_PCIDEV_BUSPROVIDER(dev) = sn_pci_provider[bs->bs_asic_type]; - } else { - SN_PCIDEV_BUSPROVIDER(dev) = &sn_pci_default_provider; - } - - /* Only set up IRQ stuff if this device has a host bus context */ - if (bs && sn_irq_info->irq_irq) { - pcidev_info->pdi_sn_irq_info = sn_irq_info; - dev->irq = pcidev_info->pdi_sn_irq_info->irq_irq; - sn_irq_fixup(dev, sn_irq_info); - } else { - pcidev_info->pdi_sn_irq_info = NULL; - kfree(sn_irq_info); - } } /* * sn_pci_controller_fixup() - This routine sets up a bus's resources - * consistent with the Linux PCI abstraction layer. + * consistent with the Linux PCI abstraction layer. */ -void sn_pci_controller_fixup(int segment, int busnum, struct pci_bus *bus) +static void +sn_pci_controller_fixup(int segment, int busnum, struct pci_bus *bus) { - int status; - int nasid, cnode; + s64 status = 0; struct pci_controller *controller; - struct sn_pci_controller *sn_controller; struct pcibus_bussoft *prom_bussoft_ptr; - struct hubdev_info *hubdev_info; - void *provider_soft; - struct sn_pcibus_provider *provider; + status = sal_get_pcibus_info((u64) segment, (u64) busnum, (u64) ia64_tpa(&prom_bussoft_ptr)); @@ -480,261 +239,77 @@ void sn_pci_controller_fixup(int segment, int busnum, struct pci_bus *bus) return; /*bus # does not exist */ prom_bussoft_ptr = __va(prom_bussoft_ptr); - /* Allocate a sn_pci_controller, which has a pci_controller struct - * as the first member. - */ - sn_controller = kzalloc(sizeof(struct sn_pci_controller), GFP_KERNEL); - if (!sn_controller) + controller = kzalloc(sizeof(*controller), GFP_KERNEL); + if (!controller) BUG(); - INIT_LIST_HEAD(&sn_controller->pcidev_info); - controller = &sn_controller->pci_controller; controller->segment = segment; - if (bus == NULL) { - bus = pci_scan_bus(busnum, &pci_root_ops, controller); - if (bus == NULL) - goto error_return; /* error, or bus already scanned */ - bus->sysdata = NULL; - } - - if (bus->sysdata) - goto error_return; /* sysdata already alloc'd */ - /* - * Per-provider fixup. Copies the contents from prom to local - * area and links SN_PCIBUS_BUSSOFT(). + * Temporarily save the prom_bussoft_ptr for use by sn_bus_fixup(). + * (platform_data will be overwritten later in sn_common_bus_fixup()) */ + controller->platform_data = prom_bussoft_ptr; - if (prom_bussoft_ptr->bs_asic_type >= PCIIO_ASIC_MAX_TYPES) - goto error_return; /* unsupported asic type */ - - if (prom_bussoft_ptr->bs_asic_type == PCIIO_ASIC_TYPE_PPB) - goto error_return; /* no further fixup necessary */ - - provider = sn_pci_provider[prom_bussoft_ptr->bs_asic_type]; - if (provider == NULL) - goto error_return; /* no provider registerd for this asic */ + bus = pci_scan_bus(busnum, &pci_root_ops, controller); + if (bus == NULL) + goto error_return; /* error, or bus already scanned */ bus->sysdata = controller; - if (provider->bus_fixup) - provider_soft = (*provider->bus_fixup) (prom_bussoft_ptr, controller); - else - provider_soft = NULL; - - if (provider_soft == NULL) { - /* fixup failed or not applicable */ - bus->sysdata = NULL; - goto error_return; - } - - /* - * Setup pci_windows for legacy IO and MEM space. - * (Temporary until ACPI support is in place.) - */ - controller->window = kcalloc(2, sizeof(struct pci_window), GFP_KERNEL); - if (controller->window == NULL) - BUG(); - controller->window[0].offset = prom_bussoft_ptr->bs_legacy_io; - controller->window[0].resource.name = "legacy_io"; - controller->window[0].resource.flags = IORESOURCE_IO; - controller->window[0].resource.start = prom_bussoft_ptr->bs_legacy_io; - controller->window[0].resource.end = - controller->window[0].resource.start + 0xffff; - controller->window[0].resource.parent = &ioport_resource; - controller->window[1].offset = prom_bussoft_ptr->bs_legacy_mem; - controller->window[1].resource.name = "legacy_mem"; - controller->window[1].resource.flags = IORESOURCE_MEM; - controller->window[1].resource.start = prom_bussoft_ptr->bs_legacy_mem; - controller->window[1].resource.end = - controller->window[1].resource.start + (1024 * 1024) - 1; - controller->window[1].resource.parent = &iomem_resource; - controller->windows = 2; - - /* - * Generic bus fixup goes here. Don't reference prom_bussoft_ptr - * after this point. - */ - - PCI_CONTROLLER(bus)->platform_data = provider_soft; - nasid = NASID_GET(SN_PCIBUS_BUSSOFT(bus)->bs_base); - cnode = nasid_to_cnodeid(nasid); - hubdev_info = (struct hubdev_info *)(NODEPDA(cnode)->pdinfo); - SN_PCIBUS_BUSSOFT(bus)->bs_xwidget_info = - &(hubdev_info->hdi_xwidget_info[SN_PCIBUS_BUSSOFT(bus)->bs_xid]); - /* - * If the node information we obtained during the fixup phase is invalid - * then set controller->node to -1 (undetermined) - */ - if (controller->node >= num_online_nodes()) { - struct pcibus_bussoft *b = SN_PCIBUS_BUSSOFT(bus); - - printk(KERN_WARNING "Device ASIC=%u XID=%u PBUSNUM=%u" - "L_IO=%lx L_MEM=%lx BASE=%lx\n", - b->bs_asic_type, b->bs_xid, b->bs_persist_busnum, - b->bs_legacy_io, b->bs_legacy_mem, b->bs_base); - printk(KERN_WARNING "on node %d but only %d nodes online." - "Association set to undetermined.\n", - controller->node, num_online_nodes()); - controller->node = -1; - } return; error_return: - kfree(sn_controller); + kfree(controller); return; } -void sn_bus_store_sysdata(struct pci_dev *dev) +/* + * sn_bus_fixup + */ +void +sn_bus_fixup(struct pci_bus *bus) { - struct sysdata_el *element; - - element = kzalloc(sizeof(struct sysdata_el), GFP_KERNEL); - if (!element) { - dev_dbg(dev, "%s: out of memory!\n", __FUNCTION__); - return; - } - element->sysdata = SN_PCIDEV_INFO(dev); - list_add(&element->entry, &sn_sysdata_list); -} + struct pci_dev *pci_dev = NULL; + struct pcibus_bussoft *prom_bussoft_ptr; + extern void sn_common_bus_fixup(struct pci_bus *, + struct pcibus_bussoft *); + + + if (!bus->parent) { /* If root bus */ + prom_bussoft_ptr = PCI_CONTROLLER(bus)->platform_data; + if (prom_bussoft_ptr == NULL) { + printk(KERN_ERR + "sn_bus_fixup: 0x%04x:0x%02x Unable to " + "obtain prom_bussoft_ptr\n", + pci_domain_nr(bus), bus->number); + return; + } + sn_common_bus_fixup(bus, prom_bussoft_ptr); + sn_legacy_pci_window_fixup(PCI_CONTROLLER(bus), + prom_bussoft_ptr->bs_legacy_io, + prom_bussoft_ptr->bs_legacy_mem); + } + list_for_each_entry(pci_dev, &bus->devices, bus_list) { + sn_pci_fixup_slot(pci_dev); + } -void sn_bus_free_sysdata(void) -{ - struct sysdata_el *element; - struct list_head *list, *safe; - - list_for_each_safe(list, safe, &sn_sysdata_list) { - element = list_entry(list, struct sysdata_el, entry); - list_del(&element->entry); - list_del(&(((struct pcidev_info *) - (element->sysdata))->pdi_list)); - kfree(element->sysdata); - kfree(element); - } - return; } /* - * Ugly hack to get PCI setup until we have a proper ACPI namespace. + * sn_io_init - PROM does not have ACPI support to define nodes or root buses, + * so we need to do things the hard way, including initiating the + * bus scanning ourselves. */ -#define PCI_BUSES_TO_SCAN 256 - -static int __init sn_pci_init(void) +void __init sn_io_init(void) { int i, j; - struct pci_dev *pci_dev = NULL; - - if (!ia64_platform_is("sn2") || IS_RUNNING_ON_FAKE_PROM()) - return 0; - - /* - * prime sn_pci_provider[]. Individial provider init routines will - * override their respective default entries. - */ - - for (i = 0; i < PCIIO_ASIC_MAX_TYPES; i++) - sn_pci_provider[i] = &sn_pci_default_provider; - pcibr_init_provider(); - tioca_init_provider(); - tioce_init_provider(); - - /* - * This is needed to avoid bounce limit checks in the blk layer - */ - ia64_max_iommu_merge_mask = ~PAGE_MASK; sn_fixup_ionodes(); - sn_irq_lh_init(); - INIT_LIST_HEAD(&sn_sysdata_list); - sn_init_cpei_timer(); - -#ifdef CONFIG_PROC_FS - register_sn_procfs(); -#endif /* busses are not known yet ... */ for (i = 0; i <= max_segment_number; i++) for (j = 0; j <= max_pcibus_number; j++) sn_pci_controller_fixup(i, j, NULL); - - /* - * Generic Linux PCI Layer has created the pci_bus and pci_dev - * structures - time for us to add our SN PLatform specific - * information. - */ - - while ((pci_dev = - pci_get_device(PCI_ANY_ID, PCI_ANY_ID, pci_dev)) != NULL) - sn_pci_fixup_slot(pci_dev); - - sn_ioif_inited = 1; /* sn I/O infrastructure now initialized */ - - return 0; -} - -/* - * hubdev_init_node() - Creates the HUB data structure and link them to it's - * own NODE specific data area. - */ -void hubdev_init_node(nodepda_t * npda, cnodeid_t node) -{ - struct hubdev_info *hubdev_info; - int size; - pg_data_t *pg; - - size = sizeof(struct hubdev_info); - - if (node >= num_online_nodes()) /* Headless/memless IO nodes */ - pg = NODE_DATA(0); - else - pg = NODE_DATA(node); - - hubdev_info = (struct hubdev_info *)alloc_bootmem_node(pg, size); - - npda->pdinfo = (void *)hubdev_info; } - -geoid_t -cnodeid_get_geoid(cnodeid_t cnode) -{ - struct hubdev_info *hubdev; - - hubdev = (struct hubdev_info *)(NODEPDA(cnode)->pdinfo); - return hubdev->hdi_geoid; -} - -void sn_generate_path(struct pci_bus *pci_bus, char *address) -{ - nasid_t nasid; - cnodeid_t cnode; - geoid_t geoid; - moduleid_t moduleid; - u16 bricktype; - - nasid = NASID_GET(SN_PCIBUS_BUSSOFT(pci_bus)->bs_base); - cnode = nasid_to_cnodeid(nasid); - geoid = cnodeid_get_geoid(cnode); - moduleid = geo_module(geoid); - - sprintf(address, "module_%c%c%c%c%.2d", - '0'+RACK_GET_CLASS(MODULE_GET_RACK(moduleid)), - '0'+RACK_GET_GROUP(MODULE_GET_RACK(moduleid)), - '0'+RACK_GET_NUM(MODULE_GET_RACK(moduleid)), - MODULE_GET_BTCHAR(moduleid), MODULE_GET_BPOS(moduleid)); - - /* Tollhouse requires slot id to be displayed */ - bricktype = MODULE_GET_BTYPE(moduleid); - if ((bricktype == L1_BRICKTYPE_191010) || - (bricktype == L1_BRICKTYPE_1932)) - sprintf(address, "%s^%d", address, geo_slot(geoid)); -} - -subsys_initcall(sn_pci_init); -EXPORT_SYMBOL(sn_pci_fixup_slot); -EXPORT_SYMBOL(sn_pci_unfixup_slot); -EXPORT_SYMBOL(sn_pci_controller_fixup); -EXPORT_SYMBOL(sn_bus_store_sysdata); -EXPORT_SYMBOL(sn_bus_free_sysdata); -EXPORT_SYMBOL(sn_generate_path); diff --git a/arch/ia64/sn/kernel/iomv.c b/arch/ia64/sn/kernel/iomv.c index 7ce3cdad627b..4aa4f301d56d 100644 --- a/arch/ia64/sn/kernel/iomv.c +++ b/arch/ia64/sn/kernel/iomv.c @@ -3,10 +3,11 @@ * License. See the file "COPYING" in the main directory of this archive * for more details. * - * Copyright (C) 2000-2003 Silicon Graphics, Inc. All rights reserved. + * Copyright (C) 2000-2003, 2006 Silicon Graphics, Inc. All rights reserved. */ #include <linux/module.h> +#include <linux/acpi.h> #include <asm/io.h> #include <asm/delay.h> #include <asm/vga.h> @@ -15,6 +16,7 @@ #include <asm/sn/pda.h> #include <asm/sn/sn_cpuid.h> #include <asm/sn/shub_mmr.h> +#include <asm/sn/acpi.h> #define IS_LEGACY_VGA_IOPORT(p) \ (((p) >= 0x3b0 && (p) <= 0x3bb) || ((p) >= 0x3c0 && (p) <= 0x3df)) @@ -31,11 +33,14 @@ void *sn_io_addr(unsigned long port) { if (!IS_RUNNING_ON_SIMULATOR()) { if (IS_LEGACY_VGA_IOPORT(port)) - port += vga_console_iobase; + return (__ia64_mk_io_addr(port)); /* On sn2, legacy I/O ports don't point at anything */ if (port < (64 * 1024)) return NULL; - return ((void *)(port | __IA64_UNCACHED_OFFSET)); + if (SN_ACPI_BASE_SUPPORT()) + return (__ia64_mk_io_addr(port)); + else + return ((void *)(port | __IA64_UNCACHED_OFFSET)); } else { /* but the simulator uses them... */ unsigned long addr; diff --git a/arch/ia64/sn/kernel/irq.c b/arch/ia64/sn/kernel/irq.c index 0b49459a878a..8c5bee01eaa2 100644 --- a/arch/ia64/sn/kernel/irq.c +++ b/arch/ia64/sn/kernel/irq.c @@ -117,7 +117,10 @@ struct sn_irq_info *sn_retarget_vector(struct sn_irq_info *sn_irq_info, nasid_t nasid, int slice) { int vector; + int cpuid; +#ifdef CONFIG_SMP int cpuphys; +#endif int64_t bridge; int local_widget, status; nasid_t local_nasid; @@ -146,7 +149,6 @@ struct sn_irq_info *sn_retarget_vector(struct sn_irq_info *sn_irq_info, vector = sn_irq_info->irq_irq; /* Free the old PROM new_irq_info structure */ sn_intr_free(local_nasid, local_widget, new_irq_info); - /* Update kernels new_irq_info with new target info */ unregister_intr_pda(new_irq_info); /* allocate a new PROM new_irq_info struct */ @@ -160,8 +162,10 @@ struct sn_irq_info *sn_retarget_vector(struct sn_irq_info *sn_irq_info, return NULL; } - cpuphys = nasid_slice_to_cpuid(nasid, slice); - new_irq_info->irq_cpuid = cpuphys; + /* Update kernels new_irq_info with new target info */ + cpuid = nasid_slice_to_cpuid(new_irq_info->irq_nasid, + new_irq_info->irq_slice); + new_irq_info->irq_cpuid = cpuid; register_intr_pda(new_irq_info); pci_provider = sn_pci_provider[new_irq_info->irq_bridge_type]; @@ -180,6 +184,7 @@ struct sn_irq_info *sn_retarget_vector(struct sn_irq_info *sn_irq_info, call_rcu(&sn_irq_info->rcu, sn_irq_info_free); #ifdef CONFIG_SMP + cpuphys = cpu_physical_id(cpuid); set_irq_affinity_info((vector & 0xff), cpuphys, 0); #endif @@ -299,6 +304,9 @@ void sn_irq_fixup(struct pci_dev *pci_dev, struct sn_irq_info *sn_irq_info) nasid_t nasid = sn_irq_info->irq_nasid; int slice = sn_irq_info->irq_slice; int cpu = nasid_slice_to_cpuid(nasid, slice); +#ifdef CONFIG_SMP + int cpuphys; +#endif pci_dev_get(pci_dev); sn_irq_info->irq_cpuid = cpu; @@ -311,6 +319,10 @@ void sn_irq_fixup(struct pci_dev *pci_dev, struct sn_irq_info *sn_irq_info) spin_unlock(&sn_irq_info_lock); register_intr_pda(sn_irq_info); +#ifdef CONFIG_SMP + cpuphys = cpu_physical_id(cpu); + set_irq_affinity_info(sn_irq_info->irq_irq, cpuphys, 0); +#endif } void sn_irq_unfixup(struct pci_dev *pci_dev) diff --git a/arch/ia64/sn/kernel/msi_sn.c b/arch/ia64/sn/kernel/msi_sn.c index 6ffd1f850d41..b3a435fd70fb 100644 --- a/arch/ia64/sn/kernel/msi_sn.c +++ b/arch/ia64/sn/kernel/msi_sn.c @@ -136,10 +136,6 @@ int sn_setup_msi_irq(unsigned int irq, struct pci_dev *pdev) */ msg.data = 0x100 + irq; -#ifdef CONFIG_SMP - set_irq_affinity_info(irq, sn_irq_info->irq_cpuid, 0); -#endif - write_msi_msg(irq, &msg); set_irq_chip_and_handler(irq, &sn_msi_chip, handle_edge_irq); diff --git a/arch/ia64/sn/kernel/setup.c b/arch/ia64/sn/kernel/setup.c index 7a2d824c5ce3..8571e52c2efd 100644 --- a/arch/ia64/sn/kernel/setup.c +++ b/arch/ia64/sn/kernel/setup.c @@ -388,6 +388,14 @@ void __init sn_setup(char **cmdline_p) ia64_sn_plat_set_error_handling_features(); // obsolete ia64_sn_set_os_feature(OSF_MCA_SLV_TO_OS_INIT_SLV); ia64_sn_set_os_feature(OSF_FEAT_LOG_SBES); + /* + * Note: The calls to notify the PROM of ACPI and PCI Segment + * support must be done prior to acpi_load_tables(), as + * an ACPI capable PROM will rebuild the DSDT as result + * of the call. + */ + ia64_sn_set_os_feature(OSF_PCISEGMENT_ENABLE); + ia64_sn_set_os_feature(OSF_ACPI_ENABLE); #if defined(CONFIG_VT) && defined(CONFIG_VGA_CONSOLE) @@ -413,6 +421,16 @@ void __init sn_setup(char **cmdline_p) if (! vga_console_membase) sn_scan_pcdp(); + /* + * Setup legacy IO space. + * vga_console_iobase maps to PCI IO Space address 0 on the + * bus containing the VGA console. + */ + if (vga_console_iobase) { + io_space[0].mmio_base = vga_console_iobase; + io_space[0].sparse = 0; + } + if (vga_console_membase) { /* usable vga ... make tty0 the preferred default console */ if (!strstr(*cmdline_p, "console=")) @@ -562,7 +580,7 @@ void __cpuinit sn_cpu_init(void) int slice; int cnode; int i; - static int wars_have_been_checked; + static int wars_have_been_checked, set_cpu0_number; cpuid = smp_processor_id(); if (cpuid == 0 && IS_MEDUSA()) { @@ -587,8 +605,16 @@ void __cpuinit sn_cpu_init(void) /* * Don't check status. The SAL call is not supported on all PROMs * but a failure is harmless. + * Architechtuallly, cpu_init is always called twice on cpu 0. We + * should set cpu_number on cpu 0 once. */ - (void) ia64_sn_set_cpu_number(cpuid); + if (cpuid == 0) { + if (!set_cpu0_number) { + (void) ia64_sn_set_cpu_number(cpuid); + set_cpu0_number = 1; + } + } else + (void) ia64_sn_set_cpu_number(cpuid); /* * The boot cpu makes this call again after platform initialization is @@ -751,5 +777,13 @@ int sn_prom_feature_available(int id) return 0; return test_bit(id, sn_prom_features); } + +void +sn_kernel_launch_event(void) +{ + /* ignore status until we understand possible failure, if any*/ + if (ia64_sn_kernel_launch_event()) + printk(KERN_ERR "KEXEC is not supported in this PROM, Please update the PROM.\n"); +} EXPORT_SYMBOL(sn_prom_feature_available); diff --git a/arch/ia64/sn/kernel/sn2/sn_hwperf.c b/arch/ia64/sn/kernel/sn2/sn_hwperf.c index 462ea178f49a..33367996d72d 100644 --- a/arch/ia64/sn/kernel/sn2/sn_hwperf.c +++ b/arch/ia64/sn/kernel/sn2/sn_hwperf.c @@ -189,7 +189,7 @@ static void print_pci_topology(struct seq_file *s) int e; for (sz = PAGE_SIZE; sz < 16 * PAGE_SIZE; sz += PAGE_SIZE) { - if (!(p = (char *)kmalloc(sz, GFP_KERNEL))) + if (!(p = kmalloc(sz, GFP_KERNEL))) break; e = ia64_sn_ioif_get_pci_topology(__pa(p), sz); if (e == SALRET_OK) diff --git a/arch/ia64/sn/kernel/tiocx.c b/arch/ia64/sn/kernel/tiocx.c index feaf1a6e8101..493380b2c05f 100644 --- a/arch/ia64/sn/kernel/tiocx.c +++ b/arch/ia64/sn/kernel/tiocx.c @@ -552,7 +552,7 @@ static void __exit tiocx_exit(void) bus_unregister(&tiocx_bus_type); } -subsys_initcall(tiocx_init); +fs_initcall(tiocx_init); module_exit(tiocx_exit); /************************************************************************ diff --git a/arch/ia64/sn/kernel/xpc_channel.c b/arch/ia64/sn/kernel/xpc_channel.c index 1f3540826e68..c08db9c2375d 100644 --- a/arch/ia64/sn/kernel/xpc_channel.c +++ b/arch/ia64/sn/kernel/xpc_channel.c @@ -632,7 +632,7 @@ xpc_process_connect(struct xpc_channel *ch, unsigned long *irq_flags) ch->number, ch->partid); spin_unlock_irqrestore(&ch->lock, *irq_flags); - xpc_create_kthreads(ch, 1); + xpc_create_kthreads(ch, 1, 0); spin_lock_irqsave(&ch->lock, *irq_flags); } @@ -754,12 +754,12 @@ xpc_process_disconnect(struct xpc_channel *ch, unsigned long *irq_flags) /* make sure all activity has settled down first */ - if (atomic_read(&ch->references) > 0 || - ((ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) && - !(ch->flags & XPC_C_DISCONNECTINGCALLOUT_MADE))) { + if (atomic_read(&ch->kthreads_assigned) > 0 || + atomic_read(&ch->references) > 0) { return; } - DBUG_ON(atomic_read(&ch->kthreads_assigned) != 0); + DBUG_ON((ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) && + !(ch->flags & XPC_C_DISCONNECTINGCALLOUT_MADE)); if (part->act_state == XPC_P_DEACTIVATING) { /* can't proceed until the other side disengages from us */ @@ -1651,6 +1651,11 @@ xpc_disconnect_channel(const int line, struct xpc_channel *ch, /* wake all idle kthreads so they can exit */ if (atomic_read(&ch->kthreads_idle) > 0) { wake_up_all(&ch->idle_wq); + + } else if ((ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) && + !(ch->flags & XPC_C_DISCONNECTINGCALLOUT)) { + /* start a kthread that will do the xpcDisconnecting callout */ + xpc_create_kthreads(ch, 1, 1); } /* wake those waiting to allocate an entry from the local msg queue */ diff --git a/arch/ia64/sn/kernel/xpc_main.c b/arch/ia64/sn/kernel/xpc_main.c index fa96dfc0e1aa..7a387d237363 100644 --- a/arch/ia64/sn/kernel/xpc_main.c +++ b/arch/ia64/sn/kernel/xpc_main.c @@ -681,7 +681,7 @@ xpc_activate_kthreads(struct xpc_channel *ch, int needed) dev_dbg(xpc_chan, "create %d new kthreads, partid=%d, channel=%d\n", needed, ch->partid, ch->number); - xpc_create_kthreads(ch, needed); + xpc_create_kthreads(ch, needed, 0); } @@ -775,26 +775,28 @@ xpc_daemonize_kthread(void *args) xpc_kthread_waitmsgs(part, ch); } - if (atomic_dec_return(&ch->kthreads_assigned) == 0) { - spin_lock_irqsave(&ch->lock, irq_flags); - if ((ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) && - !(ch->flags & XPC_C_DISCONNECTINGCALLOUT)) { - ch->flags |= XPC_C_DISCONNECTINGCALLOUT; - spin_unlock_irqrestore(&ch->lock, irq_flags); + /* let registerer know that connection is disconnecting */ - xpc_disconnect_callout(ch, xpcDisconnecting); - - spin_lock_irqsave(&ch->lock, irq_flags); - ch->flags |= XPC_C_DISCONNECTINGCALLOUT_MADE; - } + spin_lock_irqsave(&ch->lock, irq_flags); + if ((ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) && + !(ch->flags & XPC_C_DISCONNECTINGCALLOUT)) { + ch->flags |= XPC_C_DISCONNECTINGCALLOUT; spin_unlock_irqrestore(&ch->lock, irq_flags); + + xpc_disconnect_callout(ch, xpcDisconnecting); + + spin_lock_irqsave(&ch->lock, irq_flags); + ch->flags |= XPC_C_DISCONNECTINGCALLOUT_MADE; + } + spin_unlock_irqrestore(&ch->lock, irq_flags); + + if (atomic_dec_return(&ch->kthreads_assigned) == 0) { if (atomic_dec_return(&part->nchannels_engaged) == 0) { xpc_mark_partition_disengaged(part); xpc_IPI_send_disengage(part); } } - xpc_msgqueue_deref(ch); dev_dbg(xpc_chan, "kthread exiting, partid=%d, channel=%d\n", @@ -818,7 +820,8 @@ xpc_daemonize_kthread(void *args) * partition. */ void -xpc_create_kthreads(struct xpc_channel *ch, int needed) +xpc_create_kthreads(struct xpc_channel *ch, int needed, + int ignore_disconnecting) { unsigned long irq_flags; pid_t pid; @@ -833,16 +836,38 @@ xpc_create_kthreads(struct xpc_channel *ch, int needed) * kthread. That kthread is responsible for doing the * counterpart to the following before it exits. */ + if (ignore_disconnecting) { + if (!atomic_inc_not_zero(&ch->kthreads_assigned)) { + /* kthreads assigned had gone to zero */ + BUG_ON(!(ch->flags & + XPC_C_DISCONNECTINGCALLOUT_MADE)); + break; + } + + } else if (ch->flags & XPC_C_DISCONNECTING) { + break; + + } else if (atomic_inc_return(&ch->kthreads_assigned) == 1) { + if (atomic_inc_return(&part->nchannels_engaged) == 1) + xpc_mark_partition_engaged(part); + } (void) xpc_part_ref(part); xpc_msgqueue_ref(ch); - if (atomic_inc_return(&ch->kthreads_assigned) == 1 && - atomic_inc_return(&part->nchannels_engaged) == 1) { - xpc_mark_partition_engaged(part); - } pid = kernel_thread(xpc_daemonize_kthread, (void *) args, 0); if (pid < 0) { /* the fork failed */ + + /* + * NOTE: if (ignore_disconnecting && + * !(ch->flags & XPC_C_DISCONNECTINGCALLOUT)) is true, + * then we'll deadlock if all other kthreads assigned + * to this channel are blocked in the channel's + * registerer, because the only thing that will unblock + * them is the xpcDisconnecting callout that this + * failed kernel_thread would have made. + */ + if (atomic_dec_return(&ch->kthreads_assigned) == 0 && atomic_dec_return(&part->nchannels_engaged) == 0) { xpc_mark_partition_disengaged(part); @@ -857,9 +882,6 @@ xpc_create_kthreads(struct xpc_channel *ch, int needed) * Flag this as an error only if we have an * insufficient #of kthreads for the channel * to function. - * - * No xpc_msgqueue_ref() is needed here since - * the channel mgr is doing this. */ spin_lock_irqsave(&ch->lock, irq_flags); XPC_DISCONNECT_CHANNEL(ch, xpcLackOfResources, |